-
CVSS Score
-The vulnerability stems from the parse_type_line function in torch/jit/annotations.py, which directly used Python's eval() on unvalidated input strings. The commit diff shows the replacement of eval() with a safer _eval_no_call method that inspects bytecode for function calls. This matches the CWE-77 description of command injection via unsafe evaluation. The GitHub advisory and CVE explicitly name this function as the attack vector.
| Package Name | Ecosystem | Vulnerable Versions | First Patched Version |
|---|---|---|---|
| torch | pip | <= 1.13.0 | 1.13.1 |